Capsule endoscopy is a pivotal diagnostic tool for detecting small bowel abnormalities such as ulcers, arteriovenous malformations (AVM), and other gastrointestinal disorders. However, manual review of thousands of endoscopic frames is time-consuming and error-prone. This project proposes an automated classification system for Wireless Capsule Endoscopy (WCE) images using deep learning. Utilizing the KAUHC dataset curated by King Abdulaziz University Hospital, which includes 3,301 annotated images, the system classifies frames into three categories: Normal, Ulcer, and AVM. A custom-built InceptionV3 model serves as the proposed architecture, outperforming existing models such as traditional CNNs, MobileNet, and SVM classifiers. The backend is developed using Django, and a user-friendly frontend is built with HTML, CSS, and JavaScript. Users can upload endoscopic images to receive real-time diagnostic predictions. This system aims to assist gastroenterologists in enhancing diagnostic accuracy and reducing workload.

Keywords: Capsule Endoscopy, InceptionV3, Wireless Endoscopy, AVM, Ulcer Detection, CNN, MobileNet, Django, Deep Learning, KAUHC Dataset.